Claude Cowork Launches on Mobile and Web — Anthropic Previously limited to macOS and Windows desktop apps, Claude Cowork is now rolling out to mobile and web, starting with Max subscribers. The expansion enables true cross-device continuity: tasks run in the background even when no device is online, and agents pause to request human approval via smartphone before taking sensitive actions. Anthropic also revealed that over 90% of Cowork usage is non-coding knowledge work, with business operations (33.4%) as the largest single category — upending the narrative that AI coding agents are the primary use case.

🤖 Anthropic & Claude

A Global Workspace in Language Models — Anthropic Research Anthropic’s new research paper introduces “J-space,” a set of internal neural patterns in Claude that emerge from training without explicit design. These patterns let Claude reason internally, modulate thoughts across multi-step problems, and could provide a mechanism for detecting AI misbehavior by monitoring internal processing — a significant step for both interpretability and safety research.

TeraWulf Shares Surge on $19B Anthropic AI Infrastructure Lease Deal — Yahoo Finance Anthropic has committed to a $19 billion infrastructure deal with TeraWulf, signaling aggressive long-range confidence in continued compute demand. Initial capacity comes online in the second half of 2027, with a full buildout targeted for 2028.

Bringing Claude Code and Claude Cowork to Government — Claude Blog Anthropic is extending Claude Code and Cowork to government customers, expanding the enterprise reach of its agentic platform beyond commercial sectors.

Claude Code 2.1.202 — Claude Code Changelog New release adds a “Dynamic workflow size” setting in /config for controlling agent fleet scale (small/medium/large), plus workflow.run_id and workflow.name OpenTelemetry attributes — and fixes a crash in inline Ctrl+R history search.

🦾 Open Source & Models

Hy3 — Simon Willison Tencent released Hy3, a 295-billion-parameter Mixture-of-Experts model with only 21 billion active parameters. It outperforms comparably-sized models and rivals flagship open-source systems with 2–5× the parameter count, making it highly compute-efficient. The full model is on Hugging Face and free on OpenRouter until July 21.

xAI Is Dead. Long Live SpaceXAI — Gizmodo Elon Musk’s AI division xAI has rebranded to SpaceXAI, tightening its identity with SpaceX and reinforcing Musk’s narrative that AI and space infrastructure are inseparable.

Alibaba’s AI Is a Hit, but Hard to Turn Into a Moneymaker — New York Times Alibaba’s open-source Qwen models are widely adopted and significantly cheaper than US alternatives, but converting that global popularity into profitable revenue remains an unsolved problem.

State of CLI Coding Agents, Mid-2026 — arcbjorn.com A comprehensive review finds Claude Code, Codex CLI, and Omp now roughly equivalent on code quality — all can navigate real repositories, plan across files, and produce production-shaped patches. Differentiation comes down to workflow fit: permissions handling, context management, and tool availability.

🔒 Security & Safety

The ‘First’ AI-Run Ransomware Attack Still Needed a Human — TechCrunch Last week’s headlines about a “fully autonomous” AI ransomware attack require significant nuance: a human still selected the victim, built the infrastructure, and supplied stolen credentials — the AI handled only technical execution. Security firm Sysdig separately published analysis of the attack (dubbed JADEPUFFER), documenting over 600 attack payloads and calling it the first AI-executed ransomware campaign despite the human coordination role.

Savi’s App Aims to Protect Consumers from Realistic AI Scams — TechCrunch Savi launched on iOS and Android with $7M in seed funding, targeting consumers facing AI-generated scams including convincing virtual kidnapper ransom calls. The launch reflects growing concern that defensive tools have lagged behind the rapid improvement of AI-powered social engineering attacks.

The First American Autonomous Ground Vehicles Are Fighting in Ukraine — TechCrunch Forterra has deployed more than 100 self-driving ATVs in active conflict zones in Ukraine — the first deployment of American autonomous ground vehicles in combat. The milestone sharpens ongoing policy debates about AI-enabled weapons and the degree of human oversight required in autonomous military systems.

Global Push for AI Governance Amid Warnings of ‘Catastrophic Harm’ — UN News The UN Global Dialogue on AI Governance concluded in Geneva on July 7, with international stakeholders calling for coordinated frameworks covering autonomous weapons, economic disruption, and concentration of AI power.

🏭 Infrastructure & Hardware

Nvidia’s Next-Gen AI Rack System Delayed to 2028 on Manufacturing Snags — CNBC Nvidia’s Kyber rack-scale architecture — designed to house 144 Rubin Ultra chips as a unified compute system — has slipped more than 12 months to 2028. A key circuit board manufacturing challenge is to blame, and the delay signals that Nvidia’s aggressive annual release cadence is colliding with real-world hardware constraints.

Broadcom, Apple Extend Tie-Up to 2031 With New Custom Chips — Bloomberg Apple and Broadcom have extended their ASIC chip development partnership through 2031, targeting multiple generations of Apple AI products; Apple plans to deploy advanced AI servers as early as 2027.

US Investors Will Soon Get Access to SK Hynix, Another Memory Maker Riding the AI Boom — TechCrunch SK Hynix’s expected US IPO — valued at roughly $28B and expected Friday — gives public market investors their first direct exposure to the AI memory chip boom.

Bringing PyTorch Monarch to AMD GPUs — PyTorch Blog PyTorch Monarch now supports AMD ROCm, enabling elastic, fault-tolerant distributed training that dynamically recovers from node failures without halting large-scale training jobs.

💼 Industry & Business

Big Tech Has Suddenly Flipped on the AI Jobs Wipeout Scenario — Wall Street Journal Tech CEOs are retreating from apocalyptic job-loss predictions, now positioning AI as a productivity multiplier that keeps humans employed. Whether this reflects genuine recalibration or a calculated PR move in response to public backlash is unclear.

Microsoft Is Cutting More Than 3,000 Jobs in Xbox Division — Wall Street Journal Microsoft will lay off more than 3,000 Xbox workers — roughly 20% of the division — as the company continues cost restructuring aligned with its AI-first strategy.

AI Law Startup Norm Raises $120M, Hits Unicorn Valuation — TechCrunch Norm raised a $120M Series C led by Khosla Ventures, valuing the AI legal platform at $1.2 billion.

Generated by claude-sonnet-4-6 · 2026-07-07T10:00:00Z